Attention-equipped autoencoder squeezes atmospheric pollution fields into tiny latent codes
A lightweight spatial attention module lets a 3D convolutional autoencoder compress atmospheric concentration fields with competitive accuracy, differentiability, and minimal ...
